Du Xiaoman and Hong Kong University of Science and Technology Jointly Release Xuanyuan 4.5 Financial-Enhanced Large Model

Du Xiaoman and HKUST Jointly Release Xuanyuan 4.5 Financial Large Model

Phoenix Tech Download the app for exclusive early access November 02, 2026 11:58:37 From Beijing Updated November 02, 2026 11:59:18

On November 2, 2026, at the opening ceremony of the 2026 Hong Kong FinTech Week, Du Xiaoman, together with the Hong Kong University of Science and Technology (HKUST) and the Hong Kong Generative AI Research and Development Center (HKGAI), jointly released the Xuanyuan 4.5 financial enhanced large model. Based on RSI technology, Xuanyuan 4.5 achieves a 30% improvement in effectiveness across core financial scenarios such as customer acquisition, risk control, and customer management, at only 10% of the cost of general-purpose models.

Du Xiaoman CEO Zhu Guang and HKUST Vice-President (Chief Vice-President) Guo Yike jointly light up Xuanyuan 4.5

Du Xiaoman CEO Zhu Guang said: "HKUST and HKGAI have deep accumulated expertise in artificial intelligence, computer science, and interdisciplinary research, which complements Du Xiaoman's long-accumulated financial business experience, real application scenarios, and engineering capabilities. The three parties are jointly promoting the deep application of AI in the financial sector and helping to drive the intelligent upgrade of Hong Kong's financial industry. Hong Kong is the international headquarters of Du Xiaoman Group, and we will continue to deepen scientific research cooperation with HKUST and work with HKGAI to explore international markets."

HKUST Vice-President Guo Yike said: "HKUST has always been committed to bringing cutting-edge AI research into real industry scenarios. The Xuanyuan 4.5 financial large model jointly released with Du Xiaoman is the industry's first self-evolving financial enhanced large model. We look forward to the implementation of Xuanyuan 4.5 in Hong Kong, further enhancing Hong Kong's competitiveness as an international financial center."

Du Xiaoman CEO Zhu Guang introduces the Xuanyuan 4.5 financial enhanced large model to Hong Kong SAR Chief Executive John Lee

Xuanyuan 4.5 is the first financial enhanced large model in the financial sector to apply self-evolving RSI technology to model training and agent scenario goal adaptation, and consists of two parts: a large language model and a multimodal model.

The self-developed Xuanyuan-Machine participates in agent data synthesis, supervised fine-tuning, reinforcement learning, and other stages, enabling the model to continuously adapt to financial tasks. For complex businesses requiring multi-step analysis, "Xuanyuan" combines long-chain reinforcement learning, credit standard operating procedure rewards, and online policy distillation, integrates multi-domain expert capabilities, and builds over 100 billion tokens of scenario-enhanced data in a trustworthy and compliant financial agent environment. In the financial scenarios tested, the "Xuanyuan 4.5" model achieves stronger financial scenario capabilities at approximately one-tenth the inference cost of top general-purpose models.

Hong Kong SAR Chief Executive John Lee and attending guests pose for a group photo at the Du Xiaoman Xuanyuan 4.5 launch event

Different customers have different business rhythms and funding needs at different stages. Through financial expertise and personalized insight into user needs, "Xuanyuan" can meet a user's financial needs at different times and in different situations. For example, for customers who need short-term capital turnover and only have cash flow after receiving payment, the model matches a plan accordingly: pay interest first during the loan period, and repay the principal after the payment is collected, enabling earlier insight into customer needs and more precise matching of plans for customers.

Based on Xuanyuan, Du Xiaoman has built thousands of agents and more than 5,000 skills, with average daily token usage exceeding 100 billion. In Du Xiaoman's credit business, intelligent risk control is now fully driven by agents and achieves dynamic self-evolution according to the environment; the risk strategy iteration cycle has been compressed by more than half, responding to risk changes more promptly. Short-term delinquency risk has dropped by about 50%, and the overall annualized delinquency risk of credit has been reduced by 12%.

In credit review scenarios, the review capability of AI credit reviewers exceeds that of 80% of manual credit reviewers, and customer waiting time has been shortened from 10 minutes to 2 minutes. The recovery rate of AI post-loan specialists working independently is more than 8% higher than that of traditional robots.

In customer service scenarios, compared with traditional customer service robots, customer service agents can not only determine user intent but also, like humans, query the customer service knowledge base and call tools multiple times to solve complex customer business problems. At present, agents can independently handle 80% of high-frequency issues, with complex issues handled by humans as a fallback.

In the WeCom scenario, the WeCom agent provides communication strategies and script suggestions to customer managers based on the history of conversations with customers. With one customer manager plus the WeCom agent, the number of customers that can be served increases to 20,000, equivalent to the workload of 10 customer managers previously.

The industry's first RSI-based financial enhanced large model — Xuanyuan 4.5

It is reported that this release continues the industry-university-research cooperation between the two parties. In March 2025, Du Xiaoman and HKUST established the "HKUST-Du Xiaoman Joint Laboratory for Artificial Intelligence" and worked with HKGAI to jointly develop cutting-edge large model technologies and explore the application of large models in credit risk control. In the future, Du Xiaoman will also work with HKGAI through strategic cooperation to jointly promote the application of financial enhanced large model technology in international markets.